Colin
06-15-04, 01:25 PM
Well as everyone probably knows, I'm a big 3DMark freak. Why? I dunno, I guess it's a guy thing always wanting to be better then the next person type of thing. Anyway since I'm not really doing anything (yes I know it's a Friday night but I was <a target=new class=ftalternatingbarlinklarge href="http://bbs.pcstats.com/pcstats/messageview.cfm?catid=4&threadid=17150">snowed</a> in! <img src="i/expressions/face-icon-small-tongue.gif" border="0">), I figured I'd do a quicky thread on how memory timings affect 3dmark scores.<BR><BR>Since I don't have a AMD test system at home, this will probably be more beneficial for Intel users. I would expect that the results would be similar for AMD users though. Here's how the BIOS of the Abit BG7 (i845G) looks like first.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/bios.jpg"><BR><BR>The settings we'll be focuing on are...<BR><b>- CAS Latency Time<BR>- Act to Precharge Delay<BR>- DRAM RAS# to CAS# Delay<BR>- DRAM RAS# Precharge</b><BR><BR>Using the most aggressive memory timings (as shown above) lets see what we score. Please keep in mind here the system is heavily overclocked. The reason is the faster the system, the bigger difference we'll see. The Radeon 9700 Pro was <b>not</b> overclocked though.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark1.gif"><BR><BR>Not bad, just under 17k! <BR><BR>Now we're going to change the CAS Latency from 2 to 2.5. CAS latency usually affects system performance most.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark2.gif"><BR><BR>Hmm, only a drop of 30 points, that's much less then I would have thought! Yes I did rerun this test a few times.<BR><BR>Next I'm going to set the CAS Latency back to 2 and this time I'm going to lower the <b>Act to Precharge Delay</b> to 7.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark3.gif"><BR><BR>This time, we lose about 60 points. <BR><BR>I was very surprised to see that lowering the <b>Act to Precharge Delay</b> to 7 does to 3DMark then <b>CAS Latency</b>! Now I'm going to raise <b>Act to Precharge Delay</b> back to 5 and lower <b>DRAM RAS# to CAS# Delay</b> to 3.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark4.gif"><BR><BR>Hmm. This time we lose about 150 3dmarks! I'll be honest here, I had no idea <b>DRAM RAS# to CAS# Delay</b> had this big an affect! <BR><BR>Finally let's up the <b>DRAM RAS# to CAS# Delay</b> back to 2 and lower the <b>DRAM RAS# Precharge</b> to 3.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark5.gif"><BR><BR>To be honest I'm at a lost to explain this one. <b>DRAM RAS# to CAS# Delay</b> is obviously very important for 3DMark!<BR><BR>Let's see what 3DMark score we'd get with the most conservative memory timings.<BR><BR><img src="http://members.lycos.co.uk/csmo/Forum/3dmark/3dmark6.gif"><BR><BR>With all the timings at their most conservative, we get our lowest 3DMark score.<BR><BR>Well, I was very surprised by my results to say the least! With most other programs and benchmarks, <b>CAS Latency</b> is usually the most important memory timing but as we can see, the others should not be ignored either! In fact <b>DRAM RAS# to CAS# Delay</b> seemed to the most important memory timing for 3DMark!<BR><BR>While I don't have an AMD test system at home, I would expect that they would perform similar to what we see here. Hopefully you'll see that memory timings go along way in determining your 3DMark score and I hope you don't overlook even the smallest memory tweak! <img src="i/expressions/face-icon-small-smile.gif" border="0">